The Agent Skills Directory

[PROMPT_INJECTION]: Static detectors flagged strings such as 'Ignore previous instructions' and 'You are now...' in the skill-security-auditor/SKILL.md file. Technical review confirms these are part of a reference table explaining what the security scanner is designed to detect in third-party code and do not constitute a malicious attempt to override agent behavior.
[INDIRECT_PROMPT_INJECTION]: Tools like the pr-review-expert and skill-security-auditor ingest untrusted data from external sources (e.g., git diffs or other skill files). This represents a vulnerability surface where a malicious input could attempt to influence the agent. This is classified as a low-level concern given the primary purpose of the tool is auditing such content.
[COMMAND_EXECUTION]: Several skills include Python scripts (e.g., dep_scanner.py, env_auditor.py) that perform system-level tasks. These are consistent with the documented functionality of a powerful engineering suite and do not show signs of hidden or arbitrary code execution.
[EXTERNAL_DOWNLOADS]: The autoresearch-agent documentation provides instructions for users to clone additional skills from a community repository. While this involves third-party code, it is a transparent installation instruction rather than an automated, silent download of external payloads.

engineering-advanced-skills